Video Anomaly Detection based on Foreground Objects
Qing Ye, Zihan Song · 2023
In the context of complex scenes in surveillance environments, the detection of video anomalies is often hindered. To address this issue, this paper proposes a video anomaly detection algorithm based on foreground objects. Firstly, by combining frame differencing and object detection, the algorithm complements the motion information with appearance information, effectively separating the background region and obtaining more accurate pixel-level foreground object images. Secondly, to enhance the reconstruction error of abnormal samples, an autoencoder integrated with an attention mechanism is employed to learn the most representative features of normal patterns. Finally, a reconstruction-based anomaly discrimination method is utilized for video anomaly detection. Experimental results demonstrate the effectiveness of the proposed approach in detecting abnormal frames in videos.